ChangeLog

===============================================================================
This file will no longer be updated. Please check the git log!
===============================================================================

20130911:
	* Moved the project to GitHub! (https://github.com/olofson/audiality2)
	* The cfg-all script now creates build directories inside the source
	  tree. (Doesn't matter when we're using git.)

20130909:
	* Autowiring default behavior changed: Units will now output to ANY
	  unit with inputs down the chain; not just the next unit. This means a
	  set of wtosc units followed by a panmix, will all be wired to the
	  inputs of that last unit. Previously, all but the last wtosc would be
	  wired to the voice master output by default.

20130907:
	* Removed 'phase' from the instructions list in the syntax highlighting
	  xml. (It's been a control register of the wtosc unit since way before
	  the Audiality 2 rename!)

20130524:
	* Infinite loop check disabled when issuing instructions that will be
	  fixed up later.

20130420:
	* A2WT_* symbols renamed A2_W*.

20130328:
	* Minor comment/documentation corrections.
	* a2_IncRef() renamed as a2_Retain(), to match ZeeDraw style.

===============================================================================
20130102: First release under the new name: Audiality 2, version 1.9.0!
===============================================================================

20130102:
	* Added 'filter12' 12 dB/oct resonant bp/lp/hp filter unit.
	* Added demos/tests for 'filter12' to test.a2s.
	* Units are no longer asked to process zero frames.
	* Proper linear pitch cutoff ramping in 'filter12'; coefficient ramping
	  only within fragments.

20130101:
	* Added 'limiter' unit. (Port of the old hardwired one.)
	* Added single channel configuration to 'limiter'.
	* Made quick mono conversions of EvilNoises and K2Intro.
	* Compiler checks for 'return' instructions in the wrong place, as
	  those would crash the VM. (This should never happen, so the VM has no
	  stack check.)
	* Added bank info pane to a2test.c.
	* Added 'fbdelay' unit.

20121231:
	* Cracked the "random loud sounds" bug: wtosc a2o_wavetable() wasn't
	  clearing the output buffer when the pitch went supersonic, allowing
	  old audio to leak into the mix in certain cases, depending on buffer
	  usage patterns.

20121228:
	* Added 'dbgunit' for script controlled debugging.

20121227:
	* API calls that can set the API status now reset it to A2_OK if all
	  goes well.
	* a2c_Try() macro now resets the compiler error status.
	* Added NEG VM instruction.

20121226:
	* Back to encoding wave handles as fixed point integer values on the
	  code/VM level! (We still have f20 numbers inside the instructions...)
	* A2_voice.flags now initialized in all cases.
	* Added dirty a2_InstaKillAllVoices() hack to avoid disaster if
	  unloading objects while they're in use by a realtime context.
	* Fixed legato mode (CTRL + F-keys) in a2test.c.
	* Added A2_RTERRORS flag that enables realtime safe error reporting
	  from engine contexts.

20121225:
	* Cleaned up the "implicit API ownership" logic:
		* All API calls that create new objects set the A2_APIOWNED
		  flag on the returned handles.
		* a2_Assign() and a2_Export() now clear any A2_APIOWNED flag.
	* Wave creation/uploading API changed according to the above change.
	* Banks don't count exports as dependencies, but instead make sure that
	  any exports added are listed as dependencies. (This hopefully avoids
	  a few nasty situations that can othewise occur if application code
	  manually adds exports with a2_Export()...)
	* 'run' is deprecated! Prints a compiler warning that explains why...
	* Fixed A2_WAVEPOST; it came out A2_MAXFRAG frames bigger than needed!
	* Added 'handle' field to A2_voice, so we can detach handles properly.

20121224:
	* Removed the audio/audio->Process checks in the API calls.
	  (We don't have a context if there's no driver!)
	* Added realtime error reporting mechanism for non-critical errors.
	* Added a2_KillSub() for killing all subvoices of a voice.
	* Added a2_SendSub() for sending to all subvoices of a voice.
	* Wired the API message pump to commonly used/relevant API calls;
	  a2_Now(), a2_Starta(), a2_Playa(), a2_Senda(), a2_SendSuba(),
	  a2_Kill() and a2_KillSub().
	* Added CMake Debug target flags for the test directory.
	* Reworked a2test.c for the new API.

20121223:
	* Added another lock-free FIFO for responses to the API context.
	* Message oriented internal API for the FIFOs.
	* Removed the voice groups and switched to a handle based voice API.
	* Internal voice messages are now called events; A2_event.
	* Variable size messages in the lock-free FIFOs.

20121222:
	* Cleaned up, generic driver interface, ready to transparently support
	  additional types of drivers.
	* Added driver registry, allowing applications to add drivers that can
	  be used exactly like the builtin ones.
	* Added proper "stream" audio driver for a2_Read().
	* Added a2_String() and a2_Name().
	* Added string constants to the language and engine. (A2_TSTRING)
	* Added a2_Export().
	* a2_RegisterUnit(), a2_NewBank() etc no longer export or assign; they
	  just create objects!

20121221:
	* Added a2_Assign().
	* Added A2_nametab and A2_handletab internal utilities. (utilities.c)
	* a2_CloseCompiler() now cleans up the lexer states properly. (Fixes
	  memory leaks after compile errors in certain situations.)
	* Only TK_VALUE symbols are 16.16 fixed point - all others are integer!
	* Fixed some incorrect uses of strncmp().

20121220:
	* Added compiler.h.
	* Added A2_export, for A2_bank exports only.
	* New bank/object access API:
		* a2_Get() which takes a node handle and a path string and
		  returns the handle of the corresponding object.
		* a2_GetExport() and a2_GetExportName() to list the exports of
		  objects that have them.
	* Confined A2_symbol to be a compiler internal only. Compiler grabs
	  exports directly from objects instead.
	* '_' is now a valid character in A2S symbol names!

20121219:
	* Made A2_wave a public interface, moving all of the playback
	  implementation details to units that use waves. (They all still have
	  to agree with the way the engine prepares the waves, though, as we
	  don't want waves tied to specific oscillator units!)
	* Moved wave management code and builtin waves into waves.c.
	* Removed the somewhat broken a2_CustomWave() API. (Waves are now
	  public data that's supposed to be accessible by external addon
	  oscillator units!)
	* Added type names to rchm.
	* a2_TypeName() now wired to rchm; returns NULL for "unknown!"
	* a2_WaveAlloc() renamed a2_WaveNew() and removed the length argument.
	* a2_WaveWrite() now just buffers the data is is - allocation,
	  conversion and processing is done by a2_WavePrepare().

20121218:
	* Each program entry point has its own private code buffer now.
	* Added stacked A2_coder object to keep track of where VM code goes.
	* Added A2_handleflags, and moved other flags out of the way, so we
	  can send handle flags along with the other flags internally.
	* Added a2_UnloadAll() as a quick way of decreasing the reference
	  count on all banks loaded via a2_LoadString() and a2_Load().

20121217:
	* The xinsert unit now makes a callback with (NULL, 0, 0, <userdata>)
	  when the unit is destroyed, or the callback is replaced.
	* Added lock-free handle manager with reference counting; rchm.[ch].
	* Eliminated A2_object - bank objects now use the RCHM instead.
	* Eliminated a2_UntangleBank(), a2_CheckWaves() etc. It should all rely
	  on the handle refcounting now!
	* Replaced the A2_program 'eps' array with a map to 'funcs'.

20121216:
	* Masking off the CLOSE/DESTROY flags when forwarding init flags, as
	  these are actually per-object resource management flags...!
	* Moved the core processing and VM code into core.c.

20121215:
	* Added the a2_Read() API. (Uses the "dummy" audio driver.)
	* Implemented the Run() method on the "dummy" audio driver.
	* Added A2_I24, representing the internal buffer fixed point format,
	  to A2_sampleformats.
	* Samples changed to int32_t everywhere. (Was 'int' in some places.)
	* a2_WaveWrite() etc will now return A2_BADFORMAT instead of failing
	  silently if they receive an unsupported format code.
	* Removed bogus 6 dB attenuation in gui_oscilloscope().
	* Added Audiality 2 logo to a2test.
	* Compile errors before code generation has been enabled no longer
	  result in empty programs that crash the VM. (New bug that impacted
	  voice struct defs.)
	* Added pulse1 through pulse9 waves.

20121214:
	* Updated the test program, a2test.c to the current API.
	* Implemented stereo oscilloscopes in a2test.c.
	* Removed the stereo-only master output restriction.
	* Removed a2_SelectSysDriver() and a2_SelectAudioDriver().
	* Added driver pointer arguments to a2_Open(), allowing drivers to be
	  specified directly. Specify NULL for defaults.
	* Drivers now have a separate Destroy() method; Close() is now only
	  supposed to close the device!
	* Engine states take responsibility for any state changes they make to
	  drivers; state creates driver - state destroys driver, etc.
	* Audio driver buffer format changed to 24 bits. (8.24 fixed point.)
	* JACK driver now keeps its library handle and function pointers in the
	  driver struct, allowing multiple driver instances in the process.
	* Removed nonsense arbitrary 24 dB attenuation in  a2_ProcessMaster().
	* Added A2_config API, eliminating the increasing number of arguments
	  to a2_Open() and a2_SubState(). This structure also doubles as the
	  public interface to the current configuration of a running state!
	* Split the master/substate shared parts out into A2_sharedstate.
	* Unit Initialize() method now gets the A2_config instead of the state.
	* An in-use A2_config has a pointer to the state using it.
	* Driver Open() methods now get the A2_config, instead of Audiality
	  messing directly with fields in the driver structs.

===============================================================================
20121213: ChipSound officially renamed as Audiality 2!
-------------------------------------------------------------------------------
	* Project name is "audiality2", to avoid conflicts.
	* Development branch version bumped to 1.9.0.
	* API prefixes changed from cs_/CS_ to a2_/A2_.
	* C headers go in include/Audiality2
	* Script file extension changed from .csl to .a2s.
===============================================================================

20121213:
	* Added cs_Hermite2c()/cs_Hermite2().
	* New logo, web site updates...

20121212:
	* Actually wired local functions to the correct array in all places...
	* panmix unit:
		* Panning outside the [-1, 1] range now clamps the louder
		  channel while feeding the other channel back in inverted.
		* 2 in/1 out now implements pan as a crossfade control.
	* Minor optimization in cs_SetRamp(); no need to set rr->delta there!

20121211:
	* Removed the buffer pointer arrays from units. **inputs and **outputs
	  in CS_unit are now managed by the host instead.
	* Removed the CS_voice 'input' and 'output' CS_bus pointers.
	* CS_voice now has a raw noutputs/outputs pair for the sole purpose of
	  pointing subvoices to the right place.
	* Removed the CS_structitem inputs/outputs channel index arrays, as we
	  haven't implemented explicit wiring yet anyway.
	* Implemented replacing mode for all 'panmix' variants.
	* Sensible compile errors when trying to use '*' channel wildcards
	  where there can be no inputs or outputs.

20121210:
	* Documented subvoice audio routing and the 'inline' unit.
	* Added cs_Tap() API for tapping audio out of groups, as opposed to
	  cs_Insert() which expects the callback to overwrite them buffers with
	  output audio.
	* Proper 'xinsert' unit that handles the cs_Tap() and cs_Insert() APIs.
	* Added YM3812 style waveforms; halfsine, abssine and quartersine.
	* Optimized the wtosc replacing/adding callbacks by means of inline DSP
	  code, avoiding buffer clearing and wrapper function call overhead.
	* Autowiring tracks the current chain channel count and throws a
	  compile error if there's a mismatch.

20121209:
	* Local functions now have their own, dynamically allocated array of
	  entry point structs.
	* Voices now default to having no units if there's no struct statement!
	* Compile error for inserting a unit with inputs first in a chain!
	* Documented the modular voice structures and autowiring a little.

20121208:
	* Added builtin waveforms 'pulse5' through 'pulse50'.
	* 'def's are now be exported when on the top level scope. (Use '.def'
	  to keep them from being exported from the module.)

20121207:
	* Added replacing/adding Process distinction to the unit API.

20121205:
	* Unit input counts now also support '*' for scaling to the voice
	  output channel count.
	* Unit I/O count '*' now means "match output channel count". The new
	  '>' token means that, and also tells the autowiring to target the
	  output rather than scratch buffers.
	* Voice structures using '*' now set CS_program.buffers to a negative
	  value, to tell voices to ensure that there are enough scratch buffers
	  to match the output bus.

20121204:
	* Added 'descriptor' field to CS_unit.
	* Removed the Deinitialize callback from CS_unit.
	* Turned the cs_Inline() 'insert' hack unit into a real unit.

20121202:
	* Removed the hardwired master output effects.

20121201:
	* Removed some obsolete issues from the TODO.
	* Added register name collision check in csc_AddUnit().
	* Added default (0, 1, 2, ...) unit I/O wiring for units.
	* Added 2 channel (stereo) input support to the panmix unit.

20121130:
	* Removed most errors mentioning registers. (VM registers are no longer
	  directly visible on the CS scripting level.)
	* Added the csc_Varible() rule, as a subset to csc_SimplExp(), also
	  implementing namespaces.
	* Unit namespaces working properly in the compiler.

20121129:
	* CS_unit ninputs/noutputs init moved to the host side, and the
	  corresponding arguments removed from the Initialize() callback.
	* More voice structure wiring logic moved into the compiler.
	* Default autowiring logic:
		* Any unit having inputs adds scratch buffers to the voice.
		* Last unit in chain sends to the voice output bus. (Other
		  units send to the scratch buffers, if any.)

20121128:
	* Added CS_NESTLIMIT check and error code.
	* cs_AddUnit() now wires inputs and outputs as specified by the new
	  info in the CS_structitem.

20121127:
	* Rewrote the error handling using macros to keep the contents of
	  CS_errors and the cs_errnames[] array in one place.
	* Rewrote the VM instruction/opcode declarations using macros.
	* Moved voice struct config and wiring info into the programs, so
	  cs_PopulateVoice() can just trivially build voices to specification.

20121126:
	* Fixed CS_voice.input initialization bug that caused voices to
	  randomly mix into the wrong buffers at times.
	* Added variable input/output count support to the compiler.

20121124:
	* Removed various messy debug code that should no longer be needed.
	* Added variable input/output counts support to the unit API.
	* Adapted wtosc and panmix to the new API.

20121123:
	* Implemented voice nesting level based scratch buffer management.
	* Implemented automatic voice wiring based on unit I/O configuration.

20121122:
	* Rudimentary implementation of the struct {} construct.

20121121:
	* Removed the hardwired panmix unit for voices without a struct {} def.
	* Added check for redefinition of program message handler entry points.
	* CS_FIRSTCONTROLREG wired to CS_FIXEDREGS. (No hardwired registers use
	  the write callbacks at this point!)
	* cs_VoiceFree() doesn't clear cwrite for the fixed registers, as these
	  should never be used!

20121119:
	* Solved the "late message" bug! The VM/message processing loop in
	  cs_VoiceProcess() wasn't capping s_end to the time of the next
	  message when breaking out due to the cs_VoiceProcessMessages() return
	  code. Both exits are now the same code path.
	* Cleaned up and removed various debug code.

20121117:
	* Added sub-fragment processing support to cs_VoiceProcess() etc.
	* Hardwired inline subvoice processing. (Subvoices before local units.)
	* VM timer is now bumped before subvoice and unit processing, to avoid
	  breaking the "timer = 0" notification from terminated voices.

20121116:
	* Simple hardwired buffer management system.

20121115:
	* Fixed timing bug in the RETURN instruction; now checks the timer
	  after popping the VM stack!
	* Fixed RETURN so it doesn't hang voices sleeping in END instructions.
	* Removed the temporary mixing buffer hack.

20121114:
	* Separate DUMPCODE and DUMPCODERT macros for compiler and VM dumping.
	* Moved CS_ramper and cs_Noise() into a new public csdsp.h header.
	* Added linear and Hermite interpolators for 16 bit data to csdsp.h.

20121113:
	* Fixed the quick hack 'run' construct. (Used wrong SPAWN instruction!)
	* CS_voice.cwrite[] is now cleared when voices are freed, touching only
	  the indices that have actually been used for control registers.
	* Added SPAWNR and SPAWNDR to the right case block in csc_Code().

20121105:
	* Disabled JACK library unloading. Once loaded, it stays loaded.

20121031:
	* cs_Close() now closes the audio driver before shutting down the API.

20121030:
	* Added 'wait' instruction to the compiler.
	* Fixed the WAIT instruction. (Needs to set the VM timer...)

20121029:
	* Added panmix unit.
	* Removed hardwired vol and pan registers.

20121027:
	* Fixed the RETURN (cs_VoicePop()) PC for non-interrupt returns.
	* Removed "cs" prefix from voice unit source files.

20121026:
	* Removed wtosc SPITCH and SAMPLITUDE registers.
	* Added hardwired TRANSPOSE (tr) control register.
	* Added local functions and a new CALL instruction.

20121023:
	* Added extern "C" etc to the public headers, for C++ applications.

20120915:
	* Switched to SFIFO 2.0.

20120911:
	* cs_VoiceCall() can now only call entry points of the currently
	  running program.
	* Added cs_VoiceStart() to start a program on a new voice, including
	  populating the voice with units as specified by the program.
	* Added skeleton code for struct declarations to the compiler.
	* Compiler throws in a default wtosc, unless a voice structure has been
	  declared.
	* CS_cstate renamed as CS_compiler.
	* Fixed cs_DeleteSymbol() so it doesn't screw up the symbol stack...
	* Removed remaining traces of the former hardwired control registers.

20120910:
	* Rearranged the VM instruction enumeration a bit.
	* Implemented a smarter register write tracker, that uses an array of
	  register indices along with the mask to avoid scanning the whole bit
	  mask whenever one or more registers have changed.
	* SLEEP instruction that works much like END, except execution can be
	  continued by a message handler executing a WAKE instruction.
	* Reimplemented the cs_Insert() API via an internal voice unit.

20120909:
	* Removed the RUN instructions and the ability to call one program from
	  within another. (This feature is of limited use, has seemingly dodgy
	  behavior, and now opens up a massive can of worms with the dynamic
	  voice structures!)
	* Voices now populated with units as specified by started programs.
	* Added RETURN instruction for ending message handlers, avoiding logic
	  for both that and program END in one VM instruction.
	* Removed the bogus state dependent logic that was in the code path
	  that's now in RETURN instruction. (Message handlers always run in the
	  CS_INTERRUPT state!)
	* Added approximate WAIT instruction that goes to sleep until the
	  specified subvoice has terminated. Note that in terms of ramping of
	  local control registers, WAIT behaves like END (that is, changes are
	  applied instantly), as there is no known time frame.
		NOTE:
			This is a simple hack that's only buffer accurate. Do
			not use for anything where timing is critical!

20120908:
	* Added some documentation in csunits.h.
	* Added CS_unit.registers field that allows a unit to directly access
	  the VM registers that it's wired to.
	* CS_cregisters moved to csvm.h.
	* Reduced the size of some overkill size struct fields.

20120907:
	* Filled in most of csunits.h.
	* Moved bank object cleanup into proper destructor callbacks.
	* Implemented cs_RegisterUnit().
	* Eliminated SDL.h from csinternals.h. (Only the SDL driver should ever
	  touch that!)
	* Moved some generally interesting VM internals into a public struct
	  CS_vmstate, making it available to unit implementations.
	* Replaced PHASE and PHASER instructions with a 'phase' control
	  register on the 'wtosc' oscillator. (Non-ramping!)

20120906:
	* Added CS_NOMIP for streaming and large waveforms that don't need
	  mipmapping. (Uses padding for interpolation and looping, unless the
	  pitch is too high, in which case it switches to a slightly slower
	  inner loop.)
	* Oscillators no longer get a state pointer. (They could stash one in
	  their CS_wave variant or something if they REALLY need it...)
	* The 'noise' oscillator keeps the state global RNG state in the
	  "noise" CS_wave struct.

20120905:
	* Added cs_OpenSubState().
	* Added public low level waveform management API; cs_WaveAlloc(),
	  cs_WaveWrite() and cs_WavePrepare().

20120903:
	* Fixed the newly broken RNG in cs_Noise().
	* Fixed some warnings from the debug build.
	* cs_Close() no longer closes the sysdriver before freeing pools!
	* Fixed buffer clearing when the trylock fails in the jack driver.
	* DEBUG only defined for Debug build - not Maintainer.
	* Fixed some more warnings in cstest.c in Maintainer mode.
	* Build as a shared library on all platforms by default.

20120902:
	* Added dummy audio driver. (Not yet wired; needs to either flush
	  queued events, or do actual processing in a timer driven thread.)

20120830:
	* Added cs_LastError().
	* Adapted cstest.c to the updated API.

20120819:
	* Oscillator callbacks get the CS_state they belong to.

20120818:
	* API calls take a CS_state argument as needed.

20120817:
	* cs_Noise() inline takes a plain accumulator; not a CS_state.
	* cs_Rand() wired to its own, dedicated RNG state accumulator.

20120814:
	* Removed the global CS_state.

20120705:
	* Tweaked compressor/limiter scaling and default threshold level.

20120702:
	* Fixed final stage clipping in the JACK and SDL drivers.
	* JACK driver final clipping now ACTUALLY disabled by default.

20120701:
	* Moved CS_PCPU*AVG calculations into the engine context, to avoid
	  locking in the *GlobalProps() API calls.
	* Added cs_GetMicros(), for more accurate (as in, useful) CPU load
	  tracking.
	* cs_GetTicks() never relies on SDL; always host APIs.

20120630:
	* Added Un*x and Win32 implementations of cs_GetTicks().
	* Fixed a bunch of compiler warnings with the Debug switches.

20120629:
	* Added locking calls to CS_audiodriver.
	* Limiter now using 64 bit maths and stdint.h int64_t.
	* Added cs_GetTicks().
	* Both SDL and JACK are now optional. SDL is the default, if
	  present. Should build without audio drivers!
	* Fixed SDLmain linkage for the library and cstest application.
	* Added -d (driver selection) switch to cstest.

20120625:
	* Added driver selection API. (Still sticking with the simple
	  cs_Open() call for now.)

20120624:
	* Loading libjack.so dynamically.
	* Xrun handling to prevent certain JACK implementations from
	  freezing the whole system on overload.
	* Added 'flags' fields to the CS_*driver structs for passing init
	  flags from the API calls.
	* Added GS_NOAUTOCNX flag, disabling autoconnect in the JACK driver.

20120622:
	* Implemented JACK audio driver.

20120617:
	* Hardwired 2 channel limit moved to cs_Open(); drivers should now
	  support arbitrary channel counts.

20120613:
	* Ported chipsound.c to the new driver APIs.
	* Implemented SDL audio driver.

20120516:
	* Implemented default system driver. (malloc/free wrapper)

20120515:
	* Bumped version to 0.2.0. (Core API changes.)
	* Removed ALSA MIDI support. (The real API is much more powerful
	  anyway, so it's better to implement MIDI and similar externally.)
	* Added system and audio driver interfaces.
	* Using the sysdriver rtalloc()/rtfree() calls for object types
	  that may have to be allocated from the realtime context.

20120313:
	* cstest.c:
		* Hacked a quick demo mode.
		* Messages now go into the GUI status box.
		* Changed window size to 800x480.
		* Removed the DT-42 stuff from gui.[ch].
		* Added instructions on the main screen.
	* Added K2Trance to the demo songs.

20120312:
	* Slapped on a zlib license on, including sfifo.
	* Added PulseTronic to the demo songs.
	* Added rudimentary versioning.

20120310:
	* Eliminated the SDL audio locking in cs_Now().
	* Added cfg-release and cfg-mingw-release scripts.

20120309:
	* Disabled timestamp warning messages, but not "corrections".

20120306:
	* Added simple limiter from Audiality.

20120226:
	* Fixed DETACH messages; they now spin the VM, so it actually wakes
	  up if the voice is sleeping in an END instruction.
	* Messages for undefined handlers are now correctly ignored again!

20120219:
	* cs_VoiceProcess() ignores message processing errors. (Nothing we
	  can do about them anyway, and messages to undefined handlers
	  SHOULD just be ignored!)
	* Ending voices now hangs around for pending messages, even after
	  having been detached.

20120218:
	* CSM_MESSAGE now sets the VM timer correctly, so the handler is
	  actually executed before further message processing...
	* Smarter group driver; does a 100 ms fade when getting a volume
	  message, then goes back to sleep.

20120217:
	* Voice->voice messages are now sent using the message queues.

20120216:
	* Added DEBUG and DEBUGR VM instructions.
	* Added 'wge' and 'wle' constructs.
	* END instruction will now apply any pending register changes.
	* Updated the Kate syntax highlighting xml.
	* All TK_IF* and TK_WHILE* variants eliminated. (The keywords are
	  registered along with the corresponding VM opcodes anyway.)
	* "Fixed" SEND* timing issues. (Actually just quantizing the to
	  the start of the current fragment.)
	* PHASE now adjusts for subsample accurate timing.

20120215:
	* Hardwired global state eliminated from most of the engine.
	* Messages within a sample frame are now processed (including
	  running handler VM code) in the order received, rather than
	  pushed and processed in reverse order.

20120214:
	* Added JGE and JLE instructions.
	* 'wg' and 'wl' constructs fixed, using the above.
	* Added 'jge', 'jle', 'ifge' and 'ifle' constructs.
	* Fixed cs_regnames[] for proper VM code dumping.

20120212:
	* Added 'now_fragstart' for internal message timing, to avoid
	  screwing up API timestamping when there are multiple fragments
	  per audio buffer.
	* Added locking in cs_Now() for the now_frames/now_ticks pair.
	* cs_PumpMessages() now reads from FIFO directly into new message.

20120211:
	* cs_Now() can no longer jump backwards in time!

20120210:
	* Fixed compiler bug that broke 'kill N'.
	* MIDI disabled by default. (Using EEL MIDI in Kobo II sfxtest.)
	* Added cs_SendGroup() interface.
	* Builtin csgroupdriver now handles message 1(); volume control.
	* The 'p' register is no longer inherited by spawned voices!
	* Added CS_TIMESTAMP open flag to enable manual timestamp control
	  via cs_Now() and cs_Wait(). (Late events will be adjusted to
	  play ASAP regardless!)

20111207:
	* API "autotiming" feature fixed; now makes a proper timestamp
	  test that doesn't break when the timer wraps.

20111204:
	* cs_PumpMessages() adjusts late events before injecting them.
	* API messages are now handled correctly... it seems. :-)

20111203:
	* Removed the work buffer hack.
	* Added 'sa' secondary amplitude register, that is inherited
	  similar to 'sp'.
	* Added lock-free FIFO to avoid locking in most realtime calls.
	* Added timestamped message passing system.
	* Split the asynchronous API part out into csapi.c.
	* Split banks, symbols etc out into csbank.c.

20111126:
	* Temporary hack: Always use work buffers, so 'vol' will work
	  properly.

20111125:
	* Fixed internal error return code in cs_ObjAlloc().

20110823:
	* ALSA is now detected by CMake, setting CS_ALSA_MIDI if present.
	* Using custom code instead of the non-portable strndup().
	* Special case to build a static lib for Win32.
	* Added pkg-config .pc file.